World Draughts Forum

Hi,

PyDraughts is the DXP initiator. The reason for choosing the DXP protocol is that most engines (if not all) support it, but I will check if Pydraughts supports mixed mode, so one engine using DXP and the other (like Kingsrow) using hub.

Regards,
GW

Hi, I hope you all enjoyed watching the games played during the tournament organized by Krzsyztof, as always very interesting and useful! Now to make it easier to run these tournaments I am working on a Python script so that (most) engines can play against each other automatically using PyDraughts a...

Thank you!

Regards,
GW

Hi, How does the DXP mode work? Is pydrauhts the DXP server or listener? I guess pydraughts is the DXP initiator? All engines should use a different DXP port number when running on the same machine? Will pydraughts launch the engine or is the engine supposed to be started and configured in DXP mode ...

Thanks Krzysztof and we are all looking forward to the results of the next round!

Regards,
GW

Hi Krzysztof, That is what happens with an actively developed program: options and algorithms change all the time. I fully acknowledge (and I think all other programmers with me) all the hard work you put in playing all these engines against each other, many of which do not support the hub and/or th...

FYI, it turns out GWD has been playing with up to the 5-piece EGTBs only during the first four rounds. This leads to considerable evaluation errors of easily a half to a full man if most end-nodes happen to end-up in 6-piece EGTB positions, as the neural network has not been trained on positions wit...

Hi, Thank you! At least for GWD (but I guess also for other engines) it is required it is launched from the directory they are installed in so they can find configuration files etc. By inspecting the source code this is supported for the HubEngine command through the 'cwd' argument: from draughts.en...

Hi, The first results for 1M positions generated in the above way look promising. BTW, I scale the result of the material-only search back to the interval [0.0, 1.0] using the sigmoid 1.0/(1.0 + 10**(-(search score)/((value of a man) * 2))), so two men ahead gives a 90% probability of winning the ga...

Hi Bert, Thanks for your response. Meanwhile I have been looking into the positions that were generated. What I have been doing until now is: Play a 'random' game. If I really play randomly without a search the positions quickly become unequal (which is good as I do not only want 'drawish' positions...

Hi, I am now also testing NNUE in GWD but my networks seem to converge to a drawish value: if I calculate the average of (evaluation - 0.5)^2 before I train the network I get say 0.02, and when I train the network it converges to this value. Both Mathematica and Tensorflow give similar results. Has ...

GWD uses a slightly modified version of Turbo Dambase as the GUI frontend. Turbo Dambase runs on Fedora Core within a Windows virtual machine, and communicates with the Fedora Core backend by reading files from and writing files to a virtual machine shared folder that resides on a tmpfs filesystem w...

For GWD the code generated by the Intel compiler is 20-25% faster than the gcc compiler (running on Fedora Core).

Gijsbert

As an example of the above analysis: during the last tournament at Culemborg GWD played a game against Damage. After move 41... 21-27 from Damage the following position was on the board: http://fmjd.org/dias2/save/12988019507.png GWD's position is not good, and did not improve after 48.42-38, becaus...

A good point. The example I gave 'a fail-high based on a high transposition table score, but that transposition table entry gets overwritten during the research, causing the search to fail-low' can also occur in the minimal-window searches. As GWD does use the transposition table during minimal-wind...